Electricity Fraud Detection Based on Anomaly Analysis

Authors

  • Nathalia A. da Fonseca Alves Departamento de Engenharia Elétrica, Universidade Federal da Paraíba
  • Maria Eusa Alves P. de Oliveira Departamento de Engenharia Elétrica, Universidade Federal da Paraíba
  • Juan Santos Cavalcanti de Albuquerque Departamento de Engenharia Elétrica, Universidade Federal da Paraíba
  • Héctor Raúl Chávez Arias Facultad de Ingenieria Mecánica, Universidad Nacional de Ingenieria
  • Yuri Percy Molina Rodriguez Programa de Pós-Graduação em Engenharia Elétrica, Universidade Federal da Paraíba
  • Thainá Matos Santana Delgado Programa de Pós-Graduação em Engenharia Elétrica, Universidade Federal da Paraíba
  • Camila Mara Vital Barros Programa de Pós-Graduação em Engenharia Elétrica, Universidade Federal da Paraíba

Keywords:

Machine Learning, Artificial Intelligence, Energy Theft, Decision Tree, Anomaly, Electrical Grid

Abstract

Electricity utilities face substantial financial losses due to energy theft, generating negative impacts for both consumers and the companies themselves. This article explores a methodology that uses supervised learning, namely Decision Trees, to identify anomalies in electricity distribution networks. An analysis was made of the consumption data of around 523,000 people, which was reduced to approximately 11,000 customers. These were assigned a metric classifying customers with a standard deviation of consumption above 30%. Finally, the consumers who scored the highest were effectively proven to be fraudsters after analysis of the training and test data.

Downloads

Published

2024-10-18

Issue

Vol. 4 No. 1 (2024): CBA2024

Section

Articles